<< Go back to HN Digests
Warning : This document is still a draft
Warning : This document is still a draft
2026 W11 - 03-15
First time here ? Check the introduction here
What you Missed this Week
This week, OpenSource VS AI: Reimplement anything for no deps VS Copyleft, Debian VS Redox OS policy.
Claude upgrade Sonnet and Opus to 1M context for all (+ some other new claude code features)
Yann Lecun Raising funds for its startup: $1B, an impressive number (Link)
My preference?
- Can I run AI locally (Link), let you select your setup and see which model you can run with the number of tokens per seconds.
- Chimera with LLM: copy / paste blocks to create a new LLM: Link
Articles
Model Training
- 1474 pt [Link / HN] Can I run AI locally?
- 1197 pt [Link / HN] 1M context is now generally available for Opus 4.6 and Sonnet 4.6
- 370 pt [Link / HN] BitNet: Inference framework for 1-bit LLMs
- 323 pt [Link / HN] Executing programs inside transformers with exponentially faster inference
- 85 pt [Link / HN] Tree Search Distillation for Language Models Using PPO
Learning
- 470 pt [Link / HN] Show HN: How I topped the HuggingFace open LLM leaderboard on two gaming GPUs
- 448 pt [Link / HN] LLM Architecture Gallery
- 373 pt [Link / HN] A Visual Introduction to Machine Learning (2015)
- 102 pt [Link / HN] TADA: Speech generation through text-acoustic synchronization
Coding Agents
- 1548 pt [Link / HN] Shall I implement it? No
- 479 pt [Link / HN] No, it doesn’t cost Anthropic $5k per Claude Code user
- 276 pt [Link / HN] Levels of Agentic Engineering
- 224 pt [Link / HN] Claude now creates interactive charts, diagrams and visualizations
- 278 pt [Link / HN] Many SWE-bench-Passing PRs would not be merged
- 160 pt [Link / HN] Launching the Claude Partner Network
- 83 pt [Link / HN] Code Review for Claude Code
Personal Blogs / Opinions
Investigation
- 505 pt [Link / HN] How we hacked McKinsey’s AI platform
- 271 pt [Link / HN] MCP is dead; long live MCP
- 166 pt [Link / HN] Please do not A/B test my workflow
- 154 pt [Link / HN] Document poisoning in RAG systems: How attackers corrupt AI’s sources
- 120 pt [Link / HN] Open Weights isn’t Open Training
- 102 pt [Link / HN] How much of HN is AI?
Tool for AI / made with AI
- 1421 pt [Link / HN] Malus – Clean Room as a Service
- 427 pt [Link / HN] Agents that run while I sleep
- 315 pt [Link / HN] Kotlin creator’s new language: talk to LLMs in specs, not English
- 223 pt [Link / HN] Show HN: Axe – A 12MB binary that replaces your AI framework
- 160 pt [Link / HN] Show HN: OneCLI – Vault for AI Agents in Rust
- 159 pt [Link / HN] Show HN: Klaus – OpenClaw on a VM, batteries included
- 155 pt [Link / HN] Show HN: Open-source browser for AI agents
- 151 pt [Link / HN] Run NanoClaw in Docker Sandboxes
- 144 pt [Link / HN] Show HN: Rudel – Claude Code Session Analytics
- 139 pt [Link / HN] Show HN: GitAgent – An open standard that turns any Git repo into an AI agent
- 136 pt [Link / HN] Show HN: Mcp2cli – One CLI for every API, 96-99% fewer tokens than native MCP
- 127 pt [Link / HN] Show HN: A context-aware permission guard for Claude Code
- 135 pt [Link / HN] I built a programming language using Claude Code
- 119 pt [Link / HN] Show HN: Understudy – Teach a desktop agent by demonstrating a task once
- 109 pt [Link / HN] Launch HN: Spine Swarm (YC S23) – AI agents that collaborate on a visual canvas
- 92 pt [Link / HN] Show HN: Context Gateway – Compress agent context before it hits the LLM
Optimists
- 569 pt [Link / HN] Is legal the same as legitimate: AI reimplementation and the erosion of copyleft
- 306 pt [Link / HN] Allow me to get to know you, mistakes and all
- 243 pt [Link / HN] The 100 hour gap between a vibecoded prototype and a working product
- 200 pt [Link / HN] Emacs and Vim in the Age of AI
- 114 pt [Link / HN] It’s time to move your docs in the repo
- 86 pt [Link / HN] Things I’ve Done with AI
Pessimists
- 429 pt [Link / HN] Stop Sloppypasta
- 359 pt [Link / HN] The Appalling Stupidity of Spotify’s AI DJ
- 251 pt [Link / HN] LLMs can be exhausting
- 207 pt [Link / HN] I’m 60 years old. Claude Code killed a passion
- 128 pt [Link / HN] AI didn’t simplify software engineering: It just made bad engineering easier
Other
- 108 pt [Link / HN] Reliable Software in the LLM Era
- 108 pt [Link / HN] Mathematics Distillation Challenge – Equational Theories
- 107 pt [Link / HN] Against vibes: When is a generative model useful
- 88 pt [Link / HN] We Stopped Using the Mathematics That Works
Politics / Strategy
- 611 pt [Link / HN] Yann LeCun raises $1B to build AI that understands the physical world
- 408 pt [Link / HN] Redox OS has adopted a Certificate of Origin policy and a strict no-LLM policy
- 375 pt [Link / HN] Debian decides not to decide on AI-generated contributions
- 82 pt [Link / HN] The wild six weeks for NanoClaw’s creator that led to a deal with Docker
News
- 753 pt [Link / HN] Innocent woman jailed after being misidentified using AI facial recognition
- 659 pt [Link / HN] Online age-verification tools for child safety are surveilling adults
- 423 pt [Link / HN] OpenAI is walking away from expanding its Stargate data center with Oracle
- 120 pt [Link / HN] Amazon Employees Say AI Is Just Increasing Workload
- 100 pt [Link / HN] Mark Zuckerberg creating new Applied AI engineering company, reorganises teams
Economy
- 173 pt [Link / HN] 2026 tech layoffs reach 45,000 in March
- 118 pt [Link / HN] AI should not replace people at Atlassian, says CEO
Unrelated
- 417 pt [Link / HN] The dead Internet is not a theory anymore
- 351 pt [Link / HN] How kernel anti-cheats work
- 229 pt [Link / HN] Marketing for Founders
-
169 pt [Link / HN] An opinionated take on how to do important research that matters
- 173 pt [Link / HN] A Survival Guide to a PhD (2016)
- 134 pt [Link / HN] Mesh over Bluetooth LE, TCP, or Reticulum
>> You can subscribe to my mailing list here for a monthly update. <<